visitors_a_daily = visitors[visitors['group'] == 'A'][['date', 'visitors']]
visitors_a_daily.columns = ['date', 'visitors_per_date_a']
visitors_a_cummulative = visitors_a_daily.apply(
lambda x: visitors_a_daily[visitors_a_daily['date'] <= x['date']].agg(
{'date': 'max', 'visitors_per_date_a': 'sum'}
),
axis=1,
)
visitors_a_cummulative.columns = ['date', 'visitors_cummulative_a']
visitors_b_daily = visitors[visitors['group'] == 'B'][['date', 'visitors']]
visitors_b_daily.columns = ['date', 'visitors_per_date_b']
visitors_b_cummulative = visitors_b_daily.apply(
lambda x: visitors_b_daily[visitors_b_daily['date'] <= x['date']].agg(
{'date': 'max', 'visitors_per_date_b': 'sum'}
),
axis=1,
)
visitors_b_cummulative.columns = ['date', 'visitors_cummulative_b']
orders_a_daily = (
orders[orders['group'] == 'A'][['date', 'transaction_id', 'visitor_id', 'revenue']]
.groupby('date', as_index=False)
.agg({'transaction_id': pd.Series.nunique, 'revenue': 'sum'})
)
orders_a_daily.columns = ['date', 'orders_per_date_a', 'revenue_per_date_a']
orders_a_cummulative = orders_a_daily.apply(
lambda x: orders_a_daily[orders_a_daily['date'] <= x['date']].agg(
{'date': 'max', 'orders_per_date_a': 'sum', 'revenue_per_date_a': 'sum'}
),
axis=1,
).sort_values(by=['date'])
orders_a_cummulative.columns = [
'date',
'orders_cummulative_a',
'revenue_cummulative_a',
]
orders_b_daily = (
orders[orders['group'] == 'B'][['date', 'transaction_id', 'visitor_id', 'revenue']]
.groupby('date', as_index=False)
.agg({'transaction_id': pd.Series.nunique, 'revenue': 'sum'})
)
orders_b_daily.columns = ['date', 'orders_per_date_b', 'revenue_per_date_b']
orders_b_cummulative = orders_b_daily.apply(
lambda x: orders_b_daily[orders_b_daily['date'] <= x['date']].agg(
{'date': 'max', 'orders_per_date_b': 'sum', 'revenue_per_date_b': 'sum'}
),
axis=1,
).sort_values(by=['date'])
orders_b_cummulative.columns = [
'date',
'orders_cummulative_b',
'revenue_cummulative_b',
]
data = (
orders_a_daily.merge(
orders_b_daily, left_on='date', right_on='date', how='left'
)
.merge(orders_a_cummulative, left_on='date', right_on='date', how='left')
.merge(orders_b_cummulative, left_on='date', right_on='date', how='left')
.merge(visitors_a_daily, left_on='date', right_on='date', how='left')
.merge(visitors_b_daily, left_on='date', right_on='date', how='left')
.merge(visitors_a_cummulative, left_on='date', right_on='date', how='left')
.merge(visitors_b_cummulative, left_on='date', right_on='date', how='left')
)
orders_by_users_a = (
orders[orders['group'] == 'A']
.groupby('visitor_id', as_index=False)
.agg({'transaction_id': pd.Series.nunique})
)
orders_by_users_a.columns = ['visitor_id', 'orders']
orders_by_users_b = (
orders[orders['group'] == 'B']
.groupby('visitor_id', as_index=False)
.agg({'transaction_id': pd.Series.nunique})
)
orders_by_users_b.columns = ['visitor_id', 'orders']
pd.Series(0, index=np.arange(data['visitors_per_date_a'].sum() - len(orders_by_users_a['orders'])), name='orders')
[orders_by_users_a['orders'],pd.Series(0, index=np.arange(data['visitors_per_date_a'].sum() - len(orders_by_users_a['orders'])), name='orders')]
sample_a = pd.concat([orders_by_users_a['orders'],pd.Series(0, index=np.arange(data['visitors_per_date_a'].sum() - len(orders_by_users_a['orders'])), name='orders')],axis=0)
sample_b = pd.concat([orders_by_users_b['orders'],pd.Series(0, index=np.arange(data['visitors_per_date_b'].sum() - len(orders_by_users_b['orders'])), name='orders')],axis=0)
orders_by_users_a = (
orders[orders['group'] == 'A']
.groupby('visitor_id', as_index=False)
.agg({'transaction_id': pd.Series.nunique})
)
orders_by_users_a.columns = ['visitor_id', 'orders']
orders_by_users_b = (
orders[orders['group'] == 'B']
.groupby('visitor_id', as_index=False)
.agg({'transaction_id': pd.Series.nunique})
)
orders_by_users_b.columns = ['visitor_id', 'orders']
sample_a = pd.concat(
[
orders_by_users_a['orders'],
pd.Series(
0,
index=np.arange(
data['visitors_per_date_a'].sum() - len(orders_by_users_a['orders'])
),
name='orders',
),
],
axis=0,
)
sample_b = pd.concat(
[
orders_by_users_b['orders'],
pd.Series(
0,
index=np.arange(
data['visitors_per_date_b'].sum() - len(orders_by_users_b['orders'])
),
name='orders',
),
],
axis=0,
)
print('P-value:',"{0:.3f}".format(stats.mannwhitneyu(sample_a, sample_b)[1]))
print('Относительные различия:',"{0:.3f}".format(sample_b.mean() / sample_a.mean() - 1))
